Comparison
hippo vs mem0: an open-source mem0 alternative
Both are open-source memory for AI agents. mem0 uses a language model to extract memories, and memories stored through its hosted MCP server live in your Mem0 account. Hippo keeps memories in SQLite on your machine, needs no account and no model, and wires itself into the coding agents it finds.
Both mark an older fact superseded when a newer one replaces it; mem0 does that on its hosted platform. Hippo also
lets you mark a recalled memory wrong with hippo outcome --bad, and it drops out of the top results.
Two published numbers, two different metrics
Recall@5 asks whether an answer session is in the top five results. An answer score asks whether a model answered correctly from what came back. They are not comparable, so this is not a head-to-head. Hippo's method is on the benchmarks page; mem0's figure is from its README (opens in new tab), checked 2026-09-28.
Feature by feature
Every row of the README comparison table, hippo against mem0. The Mem0 column was last checked against mem0's own pages on 2026-09-28.
| Feature | hippo | Mem0 |
|---|---|---|
| Decay by default | Yes | No |
| Retrieval strengthening | Yes | No |
| Reward-proportional decay | Yes | No |
| Hybrid search (BM25 + embeddings) | Yes | Yes (semantic + BM25 + entity) |
| Schema acceleration / knowledge graph | Yes (schema) | Partial (entity linking; graph memory on Pro) |
| Conflict detection + resolution | Yes | Partial (hosted platform marks superseded facts) |
| Multi-agent shared memory | Yes | No |
| Transfer scoring | Yes | No |
| Outcome tracking | Yes | No |
| Confidence tiers | Yes | No |
| Spatial organization | No | No |
| Lossless compression | No | No |
| Cross-tool import (ChatGPT/Claude/Cursor) | Yes | No |
| Auto-hook install | Yes | No |
| MCP server | Yes | Yes (hosted, needs an account) |
| Zero runtime deps | Yes | No |
| LongMemEval (best published) | 98.0% any / 88.5% all R@5* (local MiniLM; 99.8% any-evidence with voyage-3-large; s_cleaned, per-haystack) | 94.4 (hosted platform)** |
| Git-friendly | Yes | No |
| Framework agnostic | Yes | Partial |
| License | MIT | Apache-2.0 |
* Hippo's figures are on longmemeval_s_cleaned with a per-question haystack, each the best of five retrieval settings in the benchmark scripts, not hippo recall. Any-evidence R@5 counts a hit when any answer session is in the top 5, over all 500 questions: 98.0% with the free local MiniLM embedder (an optional install) and 99.8% with voyage-3-large (measured 2026-06-09, not re-run). All-evidence R@5 counts a hit only when every answer session is in the top 5, over the 470 questions that have an answer: 86.8 to 88.5% with MiniLM. gbrain first published 97.6%, an any-evidence score over all 500; its report (opens in new tab) now leads with all-evidence, 95.53% (449 of 470) with the paid Voyage rerank-2.5 reranker and 93.19% without it. On all-evidence recall gbrain is ahead. The June 2026 build scored 98.6 any-evidence; docs/evals/2026-09-23-longmemeval-reproduction.md (opens in new tab) has both runs. An older hippo number, 86.8% R@5 on longmemeval_oracle under pooled (non-per-haystack) retrieval, is not comparable to per-haystack figures.
** Different metric: these are end-to-end answer scores, not retrieval R@5. Mem0's 94.4 comes from its hosted platform, which its README says includes optimizations the open-source SDK lacks. Zep's 90.2% and 94.7% are accuracy figures from its homepage. Memoria's 88.78% and EverMind's 83% are overall accuracy with a reader LLM. Higher denominator + LLM helps. Not directly comparable to retrieval-only R@5 numbers above. The Mem0, Zep and Letta columns were last checked against each vendor's own pages on 2026-09-28.
What mem0 ships today (checked 2026-09-28)
-
Three ways to run it: a library (
pip install mem0aiornpm install mem0ai), a self-hosted server on Postgres with pgvector, and the hosted Mem0 platform (docs (opens in new tab)). - The open-source library needs a key for a language model provider, OpenAI by default.
- Its MCP server is hosted and needs a Mem0 Platform account and API key. In mem0's words, memories stored this way "live in your Mem0 account, not on your computer" (mem0 MCP docs (opens in new tab)).
- On the hosted platform, a background process called Dream marks an older memory superseded and links it to the fact that replaced it (Dream docs (opens in new tab)).
- Graph memory (entity linking) is listed on the Pro plan (pricing (opens in new tab)). The licence is Apache-2.0.
When to pick which
Pick hippo
The memory is for coding agents on your machine. You want no account, no API key and no model in the loop, a
store that hippo init wires into the agents you already run, lessons you can mark wrong, and
markdown mirrors you can read and commit.
Pick mem0
You are building an app whose memory belongs to your end users, the case mem0 describes as "customer support chatbots, AI assistants, and autonomous systems", and a hosted platform or a Postgres server fits your stack.
FAQ
hippo and mem0, asked directly
Is hippo an open-source mem0 alternative?
Yes, if the memory is for coding agents on your own machine. Hippo is MIT-licensed, keeps memories in SQLite in your project, needs no account or API key, and hippo init wires it into Claude Code, Codex, Cursor, OpenClaw, OpenCode and Pi. mem0 is Apache-2.0 and runs as a library, a self-hosted server or a hosted platform; its open-source library calls a language model to extract memories, OpenAI by default.
Is mem0's 94.4 on LongMemEval better than hippo's 98.0%?
Neither beats the other, because they measure different things. Hippo's 98.0% is retrieval recall@5 on LongMemEval-S with a per-question haystack: whether an answer session is in the top five results. It comes from the benchmark scripts with a free local embedder, not hippo recall. mem0's 94.4 is an end-to-end answer score: whether a model answered correctly from what came back. mem0's README says it comes from its hosted platform at a top-200 retrieval budget, with optimizations the open-source SDK lacks.